Hi,
does anybody aware of a simple approach to estimate number of biological replicates for single cell RNA sequencing ?
I am interested in a wide range of questions starting from estimation of changes in cell populations and differential gene expression to reconstruction of robust gene networks within cell cluster and between clusters.
Any idea will be highly appreciated!
Hey,
usually you should use more than 1000 cells per replicate, the more the better. I would recommend to use 3000-5000 cells to be sure.
Hope this helps.
Regards,
André
I am actually more interested to know how to estimate number of independent samples rather than number of cells depending of type of question .
in any case for thanks for taking time to answer
Hey,
sorry for the inconvenience.
The answer is nearly the same: the more the better and more reliant is your result. Usually you would use three independent samples in biological research. In my opinion you should, if you can, always aim higher and use more. Using more than 30 individual probes per sample often gives the opportunity to use standard statistics, also its often used for less samples. Furthermore the results are way better and you have less outliers or can identify those better.
Be aware, that you need to look for a correct analysis method for the intra and inter relationships.
Hope this helps a little bit better.
Regards,
André
Hi Andrey, I'm trying to answer this very question. Did you find any good resources to design scRNAseq studies? From my reading it seems that batch effects and the high cost of library construction and sequencing are the main barriers to using adequate numbers of biological replicates (i.e. many studies simply pool cells from 2-3 biological replicates, thereby avoiding batch effects). This doesn't seem like it's statistically sound, though!
Nothing too definitive. But from my experience (hate to use it as a supportive argument), my conclusion so far is not very original.
All depends on questions you ask.
Specifically,
- identifying frequent enough cell populations and finding "apple-orange" differences between treatment groups can be done with 3 biological replicates per group.
- same true for pseudo-time analyses for developmental questions
- developing more advanced analyses such as network causal inferences, etc. 3 biological replicates might be sufficient for intracellular networks , but not good enough for cell-cell interactions questions.
- finally comparing analyses of pooled vs not data it looks that pooling works well for frequent cell populations that do not vary a lot within a group (stating the obvious), more importantly when the above requirements are not fulfilled the results of pooling are misleading at best.
regarding price, it is about to become much cheaper. check the following paper : Second-Strand Synthesis-Based Massively Parallel scRNA-Seq Reveals Cellular States and Molecular Features of Human Inflammatory Skin Pathologies
- Badges
- Science topic
- Similar topics
- Biological Science
- Molecular Biology
- Biomolecules
- Nucleic Acids
- RNA